Tridi: Trilateral diffusion of 3d humans, objects, and interactions

A unified framework for generative 3D Human-Object Interaction, capable of operating in seven different modes. Built on top of the Unidiffuser formulation, our method is trained to process different combinations of input and infer the unobserved ones as a conditional denoising approach. Our method is general, covering the use cases of all previous works. At the same time, it outperforms networks specialized in individual tasks, demonstrating the benefit of a more comprehensive formulation of Human-Object Interaction.

4Deform: Neural Surface Deformation for Robust Shape Interpolation

We recover a 4D realistic deformation between two sparse unordered point clouds. After applying shape matching and registration methods to obtain rough and noisy correspondence, we leverage the theory of level-set methods to infer a deformation that is continuous and physically plausible.

Nonisotropic Gaussian Diffusion for Realistic 3D Human Motion Prediction

We propose a novel diffusion noise scheduler that takes into account the skeleton structure. Instead of assuming that every joint is an independent variable (isotropic), we use the kinematic tree to take into account the joints' interdependencies in the noise formulation (Nonisotropic). We apply this framework to the human motion prediction task, resulting in more diverse and realistic motion than previous works.

Human 3Diffusion: Realistic Avatar Creation via Explicit 3D Consistent Diffusion Models

We combine a 2D multi-view generative model with a 3DGS one to obtain a 3D generative model which, from a single RGB image, obtains general and 3D consistent results.

NICP: Neural ICP for 3D Human Registration at Scale

We propose a novel localized Neural Field (LoVD), the first self-supervised task for tuning neural fields (INT), and an efficient (takes less than a minute) scalable registration pipeline (NSR), that works with out-of-distribution data (partial point clouds, clutter, different poses, …).

Interaction Replica: Tracking human–object interaction and scene changes from human motion

Given the wearable sensors and the interactive environment, we want to estimate human and object positions and achieve visually plausible results.

CloSe: A 3D Clothing Segmentation Dataset and Model

We propose a fine-grained dataset for 3D human clothing segmentation (CloSe-D), the first learning-based 3D clothing segmentation model (CloSe-Net), and an interactive tool for refining 3D segmentation labels (CloSe-T)

NSF: Neural Surface Fields for Human Modeling from Monocular Depth

The Neural Surface Fields (NSF) defines a Neural Field on the level set of an implicit representation, providing a continuous and flexible function representation on 3D geometries. We apply it in an avatarization pipeline, learning animatable avatars with pose-dependent deformations starting from a sparse set of partial depth views.

Object pop-up: Can we infer 3D objects and their poses from human interactions alone?

We show that an unorganized 3D human point cloud provides enough information to infer a 3D interacted object, opening new directions in the human-object interaction research. We also analyze the impact of different levels of information and a saliency study about the geometrical features of the input human point cloud.

Accelerating Transformer Inference for Translation via Parallel Decoding

We propose a parallel decoding algorithm to speedup transformer inference for translation.

Reduced Representation of Deformation Fields for Effective Non-rigid Shape Matching

A novel approach for non-rigid correspondence based on mesh-free methods.

Localized Shape Modelling with Global Coherence: An Inverse Spectral Approach

We combine the spectra of different linear operators to learn of to semantically modify shape geometries.

MoMaS: Mold Manifold Simulation for Real-time Procedural Texturing

We propose a novel characterization of the mold algorithm to work on arbitrary curved surfaces.

A functional skeleton transfer

A new representation for skeleton regressors, and an efficient transfer via Laplacian eigenfunctions.

Spectral shape recovery and analysis via data-driven connections

A data-driven solution and analysis of the theoretical problem of recovering a shape from its spectrum

Shape registration in the time of transformers

We propose a transformer-based procedure for the efficient registration of non-rigid 3D point clouds.

Correspondence learning via linearly-invariant embedding

Inspired by the Functional Maps paradigma, we learn a linearly-invariant embedding for 3D Shape Matching

Farm: Functional automatic registration method for 3d human bodies

A full automatic pipeline for 3D human shapes that combine intrinsic shape matching, automatic landmarks detection, and template registration